>>Speaker 1: If the last decade of cloud was really about sort of virtualization and pay-per-use pricing mechanisms and some things around that, we're entering a new era which is an era of large scale geographically disbursed computing which has all sorts of challenges associated with it. So you know there's the CAP theorem right around consistency and availability and partition tolerance. I showed a few months ago, there's a proof I put online that for those of you that are hardcore CS-types that cloud computing demand satisfiability is NP-complete. In other words, the simple problem of having disbursed users with demand that needs to be served by disbursed nodes is actually very challenging and can't be solved perfectly and since those demands are actually time varying makes the problem even more complex. So it doesn't take just until the end of the universe to solve the problem perfectly, you know, every second you create another "till the end of the universe to solve perfectly" problem.
>>Speaker 2: Yeah, I think imbrium assumed spelling as Simon was eluding to, I mean where we really are moving into a world that, what do one call it, "hybrid cloud" I prefer like "connected cloud," but it is a world of many clouds. And so applications to have the kind of reliability and availability are going to be distributed over multiple data centers. At the same time we're seeing because of data sovereignty issues and things like that, different clouds being geographically located where they're under...in a different country for example so that you can have that. So I think that's going to change the application design as well. We brought from sort of a traditional...you web tool out all applications have been built and in the networking world, actually what sets driving is a lot of east west traffic within a data center itself as opposed to the north south traffic which smaller applications have. And now when we go the second step and we're distributing that over multiple data centers, it becomes even more important than for us to understand really what those things are and you know the laws of physics. And you've written about...actually you have a new law I think which is really talking about the latencies involved; how far away are you? So we've traditionally solved at the CDNs. We have to really start taking that into account when we're designing these applications. And I think we're also looking at the fact that it's really being dominated by data. And so that we're actually... I believe we're going to start to see... Just like there has been a lot of talk about fabric computing where we're sort of merging computing storage and networking into a single kind of fabric, that the data will be you know the dominant you know attribute there because it's going to be far easier often times to move applications than it is to move large data sets. And maybe look at genomics or any of the very, very large data sets, that's really what we're going to be about. So where the application is at any point in time is going to be actually hard to predict because there'll be multiple copies of the application running and that they're going to be geographically dispersed and even though in a human time scale, like in some of the latencies that we see, you know may be acceptable, except when you're trying to stream things. For machine to machine communication, it's going to become a real issue. And so I think we're going to see tremendous new opportunities for how do you place applications; to be close to the user, to be close to the data, and making that kind of tradeoff. Even though it may be NP-complete, but I think that now there are good enough solutions that I think that this next year we're going to see a lot of progress in that area.
>>Speaker 1: The new architecture will be a balance of both; you know all of these layers working together harmoniously and seamlessly in an intelligent fashion. You know, my challenge I think to the network service providers is to create more intelligent network capabilities with greater flexibility, accessible via an API. I'd like to see bandwidth on demand, QoS on demand, routing control, things like anycast routing, things like your network positioning system. So things that really help optimize this, especially as you look at the relative cost trends for all of these different components.
==== Transcribed by Automatic Sync Technologies ====
